CN111950589B - Point cloud region growing optimization segmentation method combined with K-means clustering - Google Patents
Point cloud region growing optimization segmentation method combined with K-means clustering Download PDFInfo
- Publication number
- CN111950589B CN111950589B CN202010634692.0A CN202010634692A CN111950589B CN 111950589 B CN111950589 B CN 111950589B CN 202010634692 A CN202010634692 A CN 202010634692A CN 111950589 B CN111950589 B CN 111950589B
- Authority
- CN
- China
- Prior art keywords
- point
- clustering
- point cloud
- centroid
- points
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a point cloud region growing optimization segmentation method combined with K-means clustering, which comprises the following steps: s1, carrying out K-means clustering on the point cloud data, obtaining the centroids of the clustered object primitives, and sequencing the centroids according to the elevations to obtain the lowest centroid point; s2, traversing all the object primitives, calculating the angle and the height difference between the centroid of each object primitive and the centroid with the lowest elevation, dividing the object primitive where the centroid meeting the height difference threshold and the angle threshold is located into ground points, and dividing the object primitives where the other centroids are located into ground object points; and S3, traversing the non-divided ground object primitives, performing region growing on the non-divided ground object primitives until all the object primitives are traversed, and ending the growing. The invention adopts the growth of the clustered object elements, can solve the problem of slow growth of the traditional three-dimensional data area, improves the operation speed of the point cloud data, and solves the problem of low segmentation precision caused by scattered point cloud segmentation results.
Description
Technical Field
The invention relates to the technical field of geospatial information systems, in particular to a point cloud region growing optimization segmentation method combined with K-means clustering.
Background
An airborne laser radar (LiDAR) system is a remote sensing system for acquiring three-dimensional spatial information of a terrain surface. The airborne laser radar measurement technology can observe the earth all day long and all weather, quickly and accurately acquire the three-dimensional coordinate data of the earth surface, and is not influenced by the external environment. In recent years, the technology has been widely used for acquiring three-dimensional point cloud data, wherein point cloud segmentation is a key link in a three-dimensional point cloud data processing technology, plays an important role in positioning and identifying a target ground object, planning a reasonable path, reconstructing a large building and the like, and is an important link in three-dimensional point cloud data scene analysis.
Point cloud segmentation is to segment a point cloud into a plurality of independent areas with similar attributes, and the point clouds in the areas all have similar properties. The region growing method has the advantages of simple principle, easy realization and the like, and is widely applied to point cloud segmentation.
Although the traditional region growing method has the advantages of simple growing idea, easy realization of algorithm and detailed point cloud segmentation, the following problems exist: the traditional region growing method grows by taking points as a unit, the point clouds are regarded as a point set, and the point sets are processed one by one until all the points in the point set are processed, so that the growing speed is too slow, and a large amount of running time and storage space are consumed. Secondly, growing by taking the point as a unit can lead all the points to be organically changed into seed points, but can also cause the problem of over-scattered segmentation, so that the segmentation precision is low and the influence of noise is large.
Disclosure of Invention
The invention aims to provide a point cloud region growing and optimizing segmentation method combined with K-means clustering, and aims to solve the problems of too low growing speed and low segmentation precision in the prior art.
A point cloud region growing optimization segmentation method combined with K-means clustering comprises the following steps:
s1, performing K-means clustering on the point cloud data, acquiring the centroids of the clustered object primitives, and sequencing the centroids according to the elevations to acquire the lowest centroid point;
s2, traversing all the object primitives, calculating the angle and the height difference between the centroid of each object primitive and the centroid with the lowest elevation, dividing the object primitive where the centroid meeting the height difference threshold and the angle threshold is located into ground points, and dividing the object primitives where the other centroids are located into ground object points;
and S3, traversing the undivided ground object primitives, performing region growing on the undivided ground object primitives until all the object primitives are traversed, and ending the growing.
According to the method for optimally dividing the point cloud region by combining K-means clustering provided by the invention, the mode that the traditional region growing algorithm grows by taking points as units is changed, and the clustered object elements are grown, so that the problem that the traditional three-dimensional data region grows too slowly can be solved, the operation speed of point cloud data is improved, and the problem that the dividing precision is too low due to the scattered point cloud dividing results is solved.
The method integrates partial characteristics of a K-means algorithm and a region growing segmentation method, the K-means algorithm has the advantage of high classification speed, the region growing method has good robustness, and the segmentation of more complex point clouds with larger data volume can be realized. The actual measurement result shows that the method provided by the invention is high in segmentation speed and more stable in segmentation performance, and the experimental analysis is carried out by adopting 3 groups of point cloud data of different regions, so that the segmentation precision of the method provided by the invention can reach 82.86%, and the accuracy is greatly improved compared with the accuracy of airborne LiDAR point cloud segmentation of the traditional K-means clustering method and the region growing method. The method provided by the invention has the advantages that the processing time is about 3 minutes when the point cloud data of hundreds of thousands of orders of magnitude are processed, and the operation efficiency can be obviously improved compared with the traditional region growing method.
In addition, the point cloud region growing optimization segmentation method combined with K-means clustering can also have the following additional technical characteristics:
further, in step S1, the step of obtaining the centroid of each clustered object primitive specifically includes:
s11, randomly selecting K points from the point cloud, namely K-means clustering to obtain an initial clustering center, and dividing the point cloud data into K clustering areas;
s12, calculating the distance between the rest points and k clustering centers, and clustering the non-divided points to the clustering area where the clustering center with the nearest distance is located;
s13, calculating the average value of the point cloud data in each clustered area, and taking the average value as a new clustering center, namely updating the position of the clustering center;
and S14, performing iterative calculation until the positions of the clustering centers of all the regions are not changed any more, outputting a partitioning result, and finishing clustering.
Further, in step S1, the coordinates of the centroid point after each object primitive is clustered are calculated using the following formula:
in the formula (X) i ,Y i ,Z i ) Is P i Three-dimensional coordinates of points, Obj i For the ith object cell, the number of objects,
k is the centroid point of the object primitive, and k is the number of the object primitives.
Further, step S2 specifically includes:
traversing all centroids, setting a height difference threshold value H _ th and an angle threshold value theta _ th by calculating the height difference H and the three-dimensional angle theta between the centroid of each object primitive and the centroid with the minimum elevation, if the conditions are met, dividing the area where the centroids meeting the conditions into ground points, otherwise, dividing the area into overgrowth-free ground object areas, wherein the centroid point P (x) is the centroid point P (x is the centroid point X) i ,y i ,z i ) With the lowest point centroid M (x) j ,y j ,z j ) The solid angle θ between the two points is calculated by the following formula:
if theta is larger than the set angle threshold value, filtering out the point P (x) i ,y i ,z i ) Otherwise, the point is retained.
Further, step S3 specifically includes:
s31, acquiring the object elements after clustering segmentation, and traversing each object element;
s32, acquiring surrounding preset adjacent points by adopting a K adjacent point search method aiming at each point of the object primitive;
s33, judging whether the object primitive of the adjacent point is the same as the object primitive of the point, if so, returning to the step S32, and if not, performing the step S34;
s34, judging whether the normal vector included angle beta of the adjacent point and the point meets the threshold included angle beta _ th or not and the distance D of the adjacent point meets the distance threshold D _ th or not, if yes, growing the object element to which the adjacent point belongs and the object element to which the point belongs, traversing all the points in the current object element, and performing the step S35, otherwise, returning to the step S33;
and S35, judging whether all the object primitives are traversed or not, outputting a segmentation result if the object primitives are traversed, and returning to the step S31 if the object primitives are not traversed.
Drawings
The above and/or additional aspects and advantages of embodiments of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a general flow chart of a point cloud region growing optimization segmentation method combined with K-means clustering according to an embodiment of the present invention;
FIG. 2 is a flow chart of the steps of growing the primitive regions of an object;
fig. 3 is a schematic diagram of a segmentation result, wherein, (a) a schematic diagram of raw data Toronto 1; (b) toronto1 schematic diagram after segmentation of the present invention; (c) raw data Toronto 2; (d) toronto2 schematic diagram after segmentation of the present invention; (e) raw data Toronto 5; (f) toronto5 schematic diagram after segmentation of the present invention;
FIG. 4 is a comparison of segmentation results obtained by different segmentation methods, wherein (a) the Toronto1 segmentation is illustrated by manual pair; (b) the Toronto1 is segmented schematically by a K-means clustering algorithm; (c) the improved region growing method of the prior art is schematically illustrated by Toronto1 segmentation; (d) the Toronto1 is schematically divided;
FIG. 5 is a comparison graph of the land feature segmentation results under different segmentation methods, wherein (a) the land feature graph is segmented manually; (b) the invention is a ground feature map; (c) and (5) clustering a terrain map by using K-means.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, the method for optimally segmenting the point cloud region growth by combining K-means clustering provided by the embodiment of the present invention includes the following steps S1 to S3:
and S1, performing K-means clustering on the point cloud data, acquiring the centroids of the clustered object primitives, and sequencing the centroids according to the elevations to acquire the lowest centroid point.
The K-means is a distance-based clustering method, objects close to each other are clustered into independent compact clusters according to a criterion function, and based on the independent compact clusters, the step of obtaining the centroid of each object element after clustering specifically comprises the following steps:
s11, randomly selecting K points from the point cloud, namely obtaining an initial clustering center through K-means clustering, and dividing the point cloud data into K clustering areas;
s12, calculating the distance between the rest points and k clustering centers, and clustering the non-divided points to the clustering area where the clustering center with the nearest distance is located;
s13, calculating the average value of the point cloud data in each clustered area, and taking the average value as a new clustering center, namely updating the position of the clustering center;
and S14, iteratively calculating until the positions of the clustering centers of all the areas are not changed any more, outputting a partitioning result, and finishing clustering.
Specifically, K object primitives are obtained through a K-means clustering algorithm, and coordinates of a centroid point after each object primitive is clustered, namely a clustering center point updated after clustering can be calculated through a formula (1). The acquisition of the object primitive centroid points is an important step in the object primitive point cloud filtering, and the accuracy of the centroid point coordinates has a great influence on the filtering effect, so the parameters of the centroid point coordinates are reasonably set, and the centroid coordinates are accurately calculated.
In the formula (X) i ,Y i ,Z i ) Is P i Three-dimensional coordinates of points, Obj i For the ith object cell, the number of objects,
k is the centroid point of the object primitive, and k is the number of the object primitives. Obtaining the centroid of the object primitive is a key step in the separation of the ground point and the feature point. If the number of the centroids is too small, different objects can be divided into a class of objects, so that the object segmentation is incomplete; if the number of the centroids is too large, the algorithm running speed is affected, so that the setting of the proper number of the centroids is very important. And determining the appropriate mass center quantity according to the segmentation effect by repeatedly performing experiments aiming at different point cloud data. When K-means clustering is carried out, segmentation is carried out according to strict clustering requirements, a proper centroid parameter is set, the point clouds in each region after the K-means clustering belong to the same object, and the K-means clustering result can be over-segmented.
S2, traversing all the object primitives, calculating the angle and the height difference between the centroid of each object primitive and the centroid with the lowest elevation, dividing the object primitive where the centroid meeting the height difference threshold and the angle threshold is located into ground points, and dividing the object primitives where the other centroids are located into ground points.
Compared with other filtering algorithms based on point primitives, the filtering method based on the object primitives can provide more semantic information for subsequent filtering, but the filtering effect is greatly influenced by the segmentation result of the object primitives, so point cloud filtering needs to be performed under the condition that the accuracy of the clustered object primitives is higher.
Step S2 specifically includes:
traversing all centroids, setting a height difference threshold value H _ th and an angle threshold value theta _ th by calculating the height difference H and the three-dimensional angle theta between the centroid of each object primitive and the centroid with the minimum elevation, if the conditions are met, dividing the area where the centroids meeting the conditions into ground points, otherwise, dividing the area into overgrowth-free ground object areas, wherein the centroid point P (x) is the centroid point P (x is the centroid point X) i ,y i ,z i ) With the lowest point centroid M (x) j ,y j ,z j ) The calculation formula of the solid angle theta is shown in (2).
If theta is larger than the set angle threshold value, filtering out the point P (x) i ,y i ,z i ) Otherwise, the point is retained.
And S3, traversing the undivided ground object primitives, performing region growing on the undivided ground object primitives until all the object primitives are traversed, and ending the growing.
After the K-means cluster acquires the object primitives, traversing each object primitive, and judging the adjacent points of the points in each object. And searching K nearest points around the point by adopting a K nearest neighbor method. The method adopts default Euclidean geometric distance, searches the Euclidean geometric distance between the point data and the rest points, sorts the data according to the increasing mode of the Euclidean geometric distance, selects the point cloud data of the first k minimum Euclidean distances, and sequentially calculates the first k nearest neighbor points of the rest point cloud data. The present invention searches for a predetermined number of neighboring points around, for example, 8, that is, the parameter k is 8. For a point in an object primitive, after acquiring a neighboring point thereof, judging whether the object primitive to which the neighboring point belongs is the same as the point, if so, judging that the point is in the middle position of the object primitive, and if not, judging that the point is positioned at the edge of the object primitive or at a position close to the edge. And for the points of which the adjacent points belong to different objects, calculating a normal vector included angle beta and a distance D between the points and the adjacent points, setting a reasonable normal vector included angle threshold value beta _ th and a reasonable distance threshold value D _ th, and if the normal vector angle threshold value and the distance threshold value are simultaneously met, growing the object primitive where the adjacent points are located and the basic primitive, namely, dividing the object primitive to which the adjacent points belong and the point object primitive into one class of objects.
The region growing method is an algorithm for growing a plurality of sub-regions into a connected region, and is mainly realized by three steps of (1) selecting a proper growing seed point; (2) setting strict region growing rules; (3) the condition for stopping the region growth is set. Although the traditional region growing method has the advantages of simple growing idea, easy algorithm realization and careful point cloud segmentation, the region growing mode is to grow by taking points as units, the point clouds are regarded as a set of the points and are processed one by one until all the points in the set are processed, so that the growing speed is too slow, a large amount of running time and storage space are consumed, and the seed points are randomly selected in a segmented region, so that the segmentation performance is unstable.
The method changes the mode that the traditional region growing method grows by taking points as units, and obtains the object elements through clustering to grow. And calculating the normal quantity angle beta and the distance D. Reasonable angle and distance thresholds are set as growth criteria.
Based on this, referring to fig. 2, step S3 specifically includes:
s31, acquiring the object elements after clustering segmentation, and traversing each object element;
s32, acquiring surrounding preset adjacent points (for example, 8) by adopting a K adjacent point search method aiming at each point of the object primitive;
s33, judging whether the object primitive of the adjacent point is the same as the object primitive of the point, if yes, returning to the step S32, if not, performing the step S34;
s34, judging whether the normal vector included angle beta of the adjacent point and the point meets the threshold included angle beta _ th or not and the distance D of the adjacent point meets the distance threshold D _ th or not, if yes, growing the object element to which the adjacent point belongs and the object element to which the point belongs, traversing all the points in the current object element, and performing the step S35, otherwise, returning to the step S33;
s35, judging whether all the object elements are traversed, if so, outputting the segmentation result, otherwise, returning to the step S31.
The above method was verified as follows:
in this embodiment, the above algorithm is analyzed, tested and compared using LiDAR point cloud data provided by International Society for Photogrammetry and Remote Sensing (ISPRS) (ftp:// wg-3-4-benchmark: LVK4jvv 7mk @ ftp. ipi. uni-hand. de), and ISPRS is selected to provide data of a Toronto experimental area for experiment, wherein the three groups of data are named Toronto1, Toronto2 and Toronto5 respectively, as shown in FIG. 3. FIGS. 3(a), (c), and (e) are schematic diagrams of the un-segmented point cloud data displayed in elevation by Toronto1, Toronto2, and Toronto5, respectively, and FIGS. 3(b), (d), and (f) are schematic diagrams of the point cloud data segmented by the present invention, respectively. This experiment operation platform: windows 1064-bit operating system, processor i5-1035G1 CPU, 1.19GHz, memory 16GB, algorithm using Matlab R2018a to realize.
Toronto1 segmentation results comparative analysis
Comparing the method with a K-means method and an improved region growing method in the prior art, wherein the improved region growing method in the prior art is improved specifically for the problem that seed points are uncertain, the seed points are set through curvature, the curvature of point cloud data is sequenced, the point cloud data starts to grow from the point with the minimum curvature, namely the flattest area, and the point cloud data grows according to a growing rule.
The comparison result is shown in fig. 4, which is a schematic diagram of the segmentation result of the experiment performed on the region by the three methods shown in fig. 4, in which the Toronto1 data is used as an example for analysis, and the diagram (a) is a manual segmentation result diagram, in which the point cloud data of the whole region is manually segmented into 66 types. The K-means clustering algorithm needs to set the number of the divided regions in advance, and the correct number can be obtained from the graph (a), so the K-means clustering algorithm can set the correct number of the regions according to the graph (a), and the graph (b) can know that the traditional K-means clustering algorithm cannot distinguish ground points from ground object points, and although the correct category number is determined, the dividing effect is not ideal; in the graph (c), the improved region growing method in the prior art has a good ground point segmentation effect, but the ground feature segmentation effect is not ideal, because the growing is performed by taking points as units, the consumed time is long, and the segmentation result shows that a lot of point cloud data are not divided into correct classes, and the ground feature segmentation effect is poor; compared with the graph (b) and the graph (c), the graph (d) has better segmentation effect on the ground and the ground features, and the running speed is higher.
The method and the K-means clustering algorithm provided by the invention are adopted to compare the segmentation results of the feature points, the K-means clustering algorithm is divided into 65 areas according to the number of the acquired correct features, three features in the scene are picked for specific analysis, the two methods are respectively adopted to perform specific analysis on the positions 1, 2 and 3 in the figure 5(a), and as can be seen from the feature 1 in the figure 5(a), the clustering algorithm provided by the invention can segment two buildings, but the K-means clustering algorithm can separate the two buildings as can be seen from the figure 5 (c); for the ground object 2 in fig. 5(a), although the top of the building is slightly flawed by the method, the overall segmentation effect is better, while fig. 5(c) divides the building into 3 regions according to the minimum clustering principle, and the over-segmentation phenomenon is serious; for the partition of the 3 rd building in fig. 5(a), the K-means clustering algorithm does not divide the whole bridge into one region but 6 segments, but the bridge is divided into one class by the classification method of the present invention, but the partition is not very complete for the edge area of the bridge, because the clustered region is too large in the K-means clustering process of the present invention, the real ground point and the ground point are divided into the same region, and this problem can be solved by adjusting the number of K-means clusters.
Quantitative analysis
The method carries out quantitative analysis on the point cloud data of the Toronto1 area, adopts a K-means clustering algorithm, a region growing method improved in the literature [11] and three methods for verification, carries out analysis according to the quantity of the point clouds correctly divided in the division result and the accuracy of the division result, and respectively analyzes two division results of all the point cloud data containing ground points and the data eliminating the ground points in order to carry out reasonable comparative analysis on the point cloud data. The accuracy of the segmentation was calculated according to equation (3), and the specific segmentation results are shown in table 1.
In the formula: accuracy represents the Accuracy of the segmentation result; k represents the number of categories; TP i Representing the number of point clouds of the ith class which are correctly segmented; FN (FN) i Indicating the number of point clouds that were miscut into other categories.
TABLE 1 evaluation index values for three segmentation methods for Toronto1 data
Table 1 reflects the comparison of the accuracy of the Toronto1 data segmentation results by the three methods, and the point cloud segmentation algorithm of the present invention can obtain a better classification effect no matter whether the whole point cloud data is divided or the remaining ground feature point cloud data with ground points removed is divided, wherein the classification accuracy is 82.86% and 77.20%, and the data segmentation accuracy is less than 40% by the improved region growing method and the K-means segmentation method in the prior art. From the accuracy and the running time in table 1, the improved region growing method of the present invention has better effect in both precision and speed, compared with the improved region growing method of the prior art, the present invention has greatly improved running speed and precision, when 97088 point cloud data are processed, the improved region growing method of the prior art takes 6175.621 seconds, while the present invention only needs 178.266 seconds, the accuracy is improved from 22.84% to 82.86%, compared with the K-means method, the method of the present invention has greatly improved segmentation accuracy although the speed is slower. In conclusion, the present invention is superior to the prior art improved region growing method and K-means dividing method in terms of dividing accuracy, and is intermediate between the prior art improved region growing method and K-means dividing method in terms of dividing efficiency.
In conclusion, the method fuses partial characteristics of the K-means algorithm and the region growing segmentation method, the K-means algorithm has the advantage of high classification speed, the region growing method has good robustness, and the point cloud which is more complex and has larger data volume can be segmented. The actual measurement result shows that the method provided by the invention is high in segmentation speed and stable in segmentation performance, and the experimental analysis is carried out by adopting 3 groups of point cloud data of different regions, so that the segmentation precision of the method provided by the invention can reach 82.86%, and the precision is greatly improved compared with the precision of airborne LiDAR point cloud segmentation of the traditional K-means clustering method and the region growing method. The method provided by the invention has the advantages that the processing time is about 3 minutes when hundreds of thousands of orders of point cloud data are processed, and the operation efficiency can be obviously improved compared with the traditional region growing method.
In the description of the specification, reference to the description of "one embodiment," "some embodiments," "an example," "a specific example," or "some examples" or the like means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
Claims (5)
1. A point cloud region growing optimization segmentation method combined with K-means clustering is characterized by comprising the following steps:
s1, carrying out K-means clustering on the point cloud data, obtaining the centroids of the clustered object primitives, and sequencing the centroids according to the elevations to obtain the lowest centroid point;
s2, traversing all the object primitives, calculating the angle and the height difference between the centroid of each object primitive and the centroid with the lowest elevation, dividing the object primitives where the centroids meeting the height difference threshold and the angle threshold are located into ground points, and dividing the object primitives where the other centroids are located into overgrowth points;
and S3, traversing the overgrown ground object primitives, performing region growing on the overgrown ground object primitives until all the object primitives are traversed, and ending the growth.
2. The method for point cloud region growing and optimal segmentation in combination with K-means clustering according to claim 1, wherein in step S1, the step of obtaining the centroid of each object element after clustering specifically comprises:
s11, randomly selecting K points from the point cloud, namely obtaining an initial clustering center through K-means clustering, and dividing the point cloud data into K clustering areas;
s12, calculating the distance between the rest points and k clustering centers, and clustering the non-divided points to the clustering area where the clustering center with the nearest distance is located;
s13, calculating the average value of the point cloud data in each clustered area, and taking the average value as a new clustering center, namely updating the position of the clustering center;
and S14, performing iterative calculation until the positions of the clustering centers of all the regions are not changed any more, outputting a partitioning result, and finishing clustering.
3. The method for optimizing and segmenting the point cloud region by combining K-means clustering according to claim 1, wherein in step S1, the coordinates of the centroid point after each object primitive clustering are calculated by the following formula:
in the formula
Is composed of
The three-dimensional coordinates of the points are,
is as follows
The number of the object primitives is one,
k is the centroid point of the object primitive, and k is the number of the object primitives.
4. The point cloud region growing and optimizing segmentation method combined with K-means clustering as claimed in claim 1, wherein the step S2 specifically comprises:
traversing all centroids by calculating the height difference between the centroid of each object element and the centroid with the minimum elevation
And solid angle
Setting a height difference threshold
And angle threshold
If the conditions are met, the area where the centroids meeting the conditions are located is divided into ground points, otherwise, the area is divided into overground object areas which do not grow, wherein the centroids are
Centroid of the lowest point
Angle of space
The formula is as follows:
if it is
If the angle is larger than the set angle threshold value, filtering out points
Otherwise, the point is retained.
5. The point cloud region growing and optimizing segmentation method combined with K-means clustering as claimed in claim 1, wherein the step S3 specifically comprises:
s31, acquiring ground feature object elements after clustering segmentation, and traversing each ground feature object element;
s32, acquiring surrounding preset adjacent points by adopting a K adjacent point search method aiming at each point of the object primitive;
s33, judging whether the object primitive of the adjacent point is the same as the object primitive of the point, if so, returning to the step S32, and if not, performing the step S34;
s34, judging the included angle between the adjacent point and the normal vector of the point
Whether or not to satisfy a threshold angle
And distance of adjacent point
Whether a distance threshold is met
If yes, growing the object primitive to which the adjacent point belongs and the object primitive to which the point belongs, and performing step S35 after all points in the current object primitive are traversed, otherwise, returning to step S33;
and S35, judging whether all the object primitives are traversed or not, outputting a segmentation result if the object primitives are traversed, and returning to the step S31 if the object primitives are not traversed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010634692.0A CN111950589B (en) | 2020-07-02 | 2020-07-02 | Point cloud region growing optimization segmentation method combined with K-means clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010634692.0A CN111950589B (en) | 2020-07-02 | 2020-07-02 | Point cloud region growing optimization segmentation method combined with K-means clustering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111950589A CN111950589A (en) | 2020-11-17 |
| CN111950589B true CN111950589B (en) | 2022-09-30 |
Family
ID=73337660
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010634692.0A Active CN111950589B (en) | 2020-07-02 | 2020-07-02 | Point cloud region growing optimization segmentation method combined with K-means clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111950589B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112435342B (en) * | 2020-11-24 | 2023-08-01 | 华侨大学 | Region division method and system for special-shaped curved surface |
| CN113139721B (en) * | 2021-04-16 | 2023-12-19 | 深圳市艾赛克科技有限公司 | Aggregate storage yard management system and method |
| CN115186153A (en) * | 2022-07-21 | 2022-10-14 | 中国建筑西南设计研究院有限公司 | Beam bottom net height analysis method, device and storage medium based on equidistant partition method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102520401A (en) * | 2011-12-21 | 2012-06-27 | 南京大学 | Building area extraction method based on LiDAR data |
| CN106772424A (en) * | 2016-11-22 | 2017-05-31 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | One kind hands over cross-line to determine method and apparatus |
| CN110866449A (en) * | 2019-10-21 | 2020-03-06 | 北京京东尚科信息技术有限公司 | Method and device for identifying target object in road |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101702200B (en) * | 2009-11-03 | 2012-02-29 | 武汉大学 | Automatic classification method of airborne laser radar point cloud data |
| CN103514598B (en) * | 2013-07-20 | 2016-08-10 | 北京航空航天大学 | A kind of automatic division method building SFM point cloud |
| CN105894044B (en) * | 2016-04-29 | 2017-08-25 | 武汉大学 | A kind of individual plant tree point cloud extraction method based on Vehicle-borne Laser Scanning data |
-
2020
- 2020-07-02 CN CN202010634692.0A patent/CN111950589B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102520401A (en) * | 2011-12-21 | 2012-06-27 | 南京大学 | Building area extraction method based on LiDAR data |
| CN106772424A (en) * | 2016-11-22 | 2017-05-31 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | One kind hands over cross-line to determine method and apparatus |
| CN110866449A (en) * | 2019-10-21 | 2020-03-06 | 北京京东尚科信息技术有限公司 | Method and device for identifying target object in road |
Non-Patent Citations (4)
| Title |
|---|
| 基于K-means聚类方法的三维点云模型分割;孙红岩;《计算机工程与应用 》;20060401;第42-45页 * |
| 基于多聚类的混合数据分割算法;张蓉等;《信息通信》;20180615(第06期);第42-44页 * |
| 点云模型分割及应用技术综述;刘进等;《计算机科学》;20110415(第04期);第21-24页 * |
| 采用模糊C均值方法进行激光点云分类;王书民等;《测绘通报》;20161025(第10期);第21-24页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111950589A (en) | 2020-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111950589B (en) | Point cloud region growing optimization segmentation method combined with K-means clustering | |
| CN106529469B (en) | Unmanned aerial vehicle-mounted LiDAR point cloud filtering method based on self-adaptive gradient | |
| CN111340723B (en) | Terrain-adaptive airborne LiDAR point cloud regularization thin plate spline interpolation filtering method | |
| CN110992341A (en) | Segmentation-based airborne LiDAR point cloud building extraction method | |
| CN109461207A (en) | A kind of point cloud data building singulation method and device | |
| CN115063555B (en) | Vehicle-mounted LiDAR point cloud street tree extraction method for Gaussian distribution area growth | |
| CN110443810A (en) | Point cloud plane dividing method based on quick adjacent voxel inquiry | |
| CN109934120B (en) | Step-by-step point cloud noise removing method based on space density and clustering | |
| CN110335352B (en) | Double-element multi-resolution hierarchical filtering method for airborne laser radar point cloud | |
| CN114266987A (en) | Intelligent identification method for high slope dangerous rock mass of unmanned aerial vehicle | |
| CN110765962A (en) | Plant identification and classification method based on three-dimensional point cloud contour dimension values | |
| CN111428784A (en) | Robust segmentation method for deciduous forest tree-level parameter measurement by adopting airborne laser radar | |
| CN114170149A (en) | Road geometric information extraction method based on laser point cloud | |
| WO2011085435A1 (en) | Classification process for an extracted object or terrain feature | |
| CN114463338B (en) | Automatic building laser foot point extraction method based on graph cutting and post-processing | |
| CN116258857A (en) | Outdoor tree-oriented laser point cloud segmentation and extraction method | |
| Wang et al. | An individual tree segmentation method from mobile mapping point clouds based on improved 3-D morphological analysis | |
| CN117765006A (en) | Multi-level dense crown segmentation method based on unmanned aerial vehicle image and laser point cloud | |
| AU2010200146A1 (en) | Extraction processes | |
| CN116051841A (en) | Roadside ground object multistage clustering segmentation algorithm based on vehicle-mounted LiDAR point cloud | |
| Qin et al. | A voxel-based filtering algorithm for mobile LiDAR data | |
| CN112381029B (en) | Method for extracting airborne LiDAR data building based on Euclidean distance | |
| KR20170016060A (en) | filtering and extraction of feature boundary method from terrestrial lidar data using data mining techniques and device thereof | |
| CN115410036A (en) | Automatic classification method for key element laser point clouds of high-voltage overhead transmission line | |
| CN115187979A (en) | Single-wood point cloud branch and leaf separation method based on graph theory |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |